When dealing with multiple flexes (specifically from mujoco_warp/test_data/flex/multiflex.xml), there are two issues (maybe related but not sure).
-
Running the viewer on all three different flexes in one scene is extremely slow and also blows up after a few frames uv run mjwarp-viewer mujoco_warp/test_data/flex/multiflex.xml --njmax 10000
-
When different multiple flexes are added, the performance is lower than expected. e.g. if the performance of cloth is 2000 FPS, adding a rope somewhere to the scene should result in an expected proportional drop. Instead the performance take a substantial hit (20x slower) even though the rope by itself has a high FPS, so the expected performance should maybe have been 1200-1800? This can be tested by running uv run mjwarp-testspeed mujoco_warp/test_data/flex/multiflex.xml --nworld 32 --njmax 6000 --nconmax 200 --nstep 100 and commenting out the soft-body and rope, and adding the rope back.
These issues might be related to #1264
When dealing with multiple flexes (specifically from
mujoco_warp/test_data/flex/multiflex.xml), there are two issues (maybe related but not sure).Running the viewer on all three different flexes in one scene is extremely slow and also blows up after a few frames
uv run mjwarp-viewer mujoco_warp/test_data/flex/multiflex.xml --njmax 10000When different multiple flexes are added, the performance is lower than expected. e.g. if the performance of cloth is 2000 FPS, adding a rope somewhere to the scene should result in an expected proportional drop. Instead the performance take a substantial hit (20x slower) even though the rope by itself has a high FPS, so the expected performance should maybe have been 1200-1800? This can be tested by running
uv run mjwarp-testspeed mujoco_warp/test_data/flex/multiflex.xml --nworld 32 --njmax 6000 --nconmax 200 --nstep 100and commenting out the soft-body and rope, and adding the rope back.These issues might be related to #1264